1. Voltage Reference Introduction TIPL 4501 TI Precision Labs – ADCs
Created by Luis Chioye, Art Kay
Presented by Cynthia Sosa
Hello, and welcome to the TI Precision Lab series covering voltage reference considerations for SAR ADCs. The goal of this section is to cover reference specifications, gain a deeper understanding of the SAR voltage refernece behavior, and develop methods for driving the reference input that minimize error.
This video starts with an introduction on voltage references. In this video we cover a brief overview of voltage reference specifications including initial accuracy, long term stability, temperature drift, and noise. In follow on videos we will cover topics related to driving the SAR reference input. Before we begin looking at reference specifications, lets cover some more detail on the objective of the entire reference series. .

2. SAR Reference Drive circuit
ADC performance is a function of Reference
Initial accuracy, drift, stability and noise
SAR Reference input:
Typically is not high impedance
Presents a dynamic load.
External Reference buffer often required
The SAR reference drive circuit is critical in the ADC, because all conversions are a function of the reference voltage.
Click
Since the ADC performs conversions by comparing the input voltage against the reference, the initial accuracy, drift, stability and noise of the reference voltage directly affect the accuracy of conversion results.
click
With a traditional successive approximation register (SAR) ADC architecture, the reference input is the most sensitive, as the reference pin presents a dynamic load to the reference source.  
In many cases, a reference buffer may be required to cope with the dynamic current demand at the SAR reference input.

3. Overall Objective
Design low drift, accurate and stable SAR reference drive circuit.
Reference source must support the dynamic charge/current at the SAR Ref input without affecting noise and stability performance.
Reference voltage must remain settled within the ADC resolution during each conversion clock cycle.
The reference drive circuit to the ADC must provide a low drift, very accurate DC voltage and support the dynamic
charge requirements without affecting noise and stability performance.
In this presentation, an overview of key voltage reference specifications will be provided; as well as a detailed explanation of the SAR reference input pin behavior.
The reference drive circuit must be able to cope with the reference input dynamic current demand. The reference voltage must remain stable and settled to the required N-Bit resolution during each conversion clock cycle, or linearity errors and missing code errors may occur.
The overall objective of this presentation is to provide a method to design an optimal reference drive circuit and verify its performance using TINA SPICE simulations.

4. Agenda for the voltage reference series
Reference Performance Specifications:
Initial Accuracy, Drift, Long Term Drift, and Noise
SAR REF Drive Circuit Topologies:
Standalone Reference vs. Buffered Reference
SAR ADCs with Internal Reference Buffer
SAR REF Input Overview: The Capacitive DAC (CDAC)
Build TINA REF Input Model for a SAR:
Discrete Charge Model
TI Device Specific Model
SAR REF Drive Circuit Design:
Reference Bypass Capacitor
Reference Buffer Stability and Compensation
This slide shows the agenda for the entire series on the SAR ADC reference. In this video we will cover the first topic: initial accuracy, drift, long term drift, and noise. Topics for the other videos are given below. Without further delay, let’s get started on reference specifications.

5. Series or Shunt Reference
Cannot sink current
Limited ability to respond to load transients
Used for wide Vin and Vin transients
Can be used as floating reference.
Used in ADC reference input drive
Output amplifier can source and sink current
Low dropout
No external resistor required.
Before starting to review different reference specifications we should cover the main differences between the two different types of voltage references: the shunt and series reference. The shunt reference acts very similar to a Zener diode. Current through the shunt reference is set by a series resistance. The shunt reference does not have an amplifier buffered so it cannot sink current. The amount of current it can source is related to the shunt resistor. Because of their limited ability to respond to load transients, these references are generally not used when driving switched capacitor reference inputs. One possible exception would be cases where the shunt reference is followed by a wide bandwidth buffer.
The series reference, on the other hand, is the type that we will be focusing on in this presentation because they are the most common type used for ADC input drive. The series regulator has an internal bandgap reference and an amplifier that buffers the reference voltage. Normally, the series regulator output can source or sink current which makes this type of regulator good for responding to transient current demands. Later in this seminar we will learn about characteristics that make some series references better than others for driving references. For now suffice it to say that we will be focusing on series voltage reference the remainder of this presentation.

6. Initial Accuracy Valid at the test conditions: see top of table.
At TA = 25⁰C, ILOAD = 0, CL = 1µF, and VIN = (VOUT + 0.2V) TO 18V, unless otherwise noted.
PARAMETER
TEST CONDITIONS
MIN
TYP
UNIT
VOUT
Output Voltage
REF5020
2.048 V
REF5025
2.500
REF5030
3.000
REF5040
4.096
REF5050
5.000
REF5010
10.00
Initial Accuracy: High Grade
-0.05
0.05 %
Initial Accuracy: Standard Grade
-0.1
0.1
Valid at the test conditions: see top of table.
Does not include effects from temperature drift, soldering, long term shift, or thermal hysteresis.
Initial accuracy is the error in the specified output voltage of the reference.
Click The error is only valid under the specific set of test conditions that the initial error is defined for.
Click
In this example, the initial accuracy is defined by the test conditions at the top of the page. That is, TA = 25⁰C, ILOAD = 0, CL = 1µF, and VIN = (VOUT + 0.2V) TO 18V, unless otherwise noted.
Notice that some references have multiple grade outs.
Click In this example, the high grade has 0.05% error and the low grade has 0.1% accuracy.
Click Note that the initial accuracy does not includes the effects of accuracy shifts from soldering, long term shift, or hysteresis. We will discuss these error sources in a moment.

7. Initial Accuracy and Solder Shift
The solder shift is for a specific test condition:
Example: PCB thickness, area, and soldering profile.
Other conditions (e.g. multiple reflows) could have larger shifts.
This slide shows the histogram for initial accuracy as well as the error introduced by solder shift. The solder-heat shift distribution error is caused by stress introduced in the die during the soldering process. This error is in addition to the initial accuracy error. This distribution was measured under very specific test conditions such as the PCB thickness, area, and soldering profile. Changing the soldering conditions can impact the solder shift. For example, many PCB designs will require more than one reflow for top and bottom sides of the PCB. This can make the solder-heat shift even larger.

8. Long Term Shift (stability)
At TA = 25⁰C, ILOAD = 0, CL = 1µF, and VIN = (VOUT + 0.2V) TO 18V, unless otherwise noted.
PARAMETER
TEST CONDITIONS
MIN
TYP
UNIT
LONG-TERM STABILITY: VSSOP-8
0 to 1000 hours at 25⁰C
125
ppm
1000 to 2000 hours at 25⁰C 45
LONG-TERM STABILITY: MSOP-8
100 50
96 units tested.
Caused by curing of the molding compounds, metal migration, and diffusion.
The initial shift (first few months) makes up the majority of change. Due to Self curing of molding compound.
Long term shift, sometimes called long term stability, is a measurement of how the accuracy of the devices changes with time. This is sometimes called the aging of the product. All devices will have long term shifts from ageing. However, this error is especially important in references.
The table shows the typical shift of the reference output for two different packages: VSSOP-8 and MSOP-8. The table specifies the output drift from 0 to 1000hours and from 1000 hours to 2000 hours.
Click
In this example, 96 units were tested.
click
Long term shifts are caused by curing of the molding compounds, metal migration, and diffusion. In the life of a product the long term shifts are largest initially and begin to diminish over time. The reason for the large initial shift is the self curing of the molding compound in the device package. One way to minimize the effects of long term shift is to periodically calibrate the product. Thus, the long term shifts are accounted for. Another approach is to do an initial “burn-in”. This means that the product is baked at high temperature to accelerate the ageing of the product and get past the “initial shift”.

9. Temperature Drift (Box Method)
𝐷𝑟𝑖𝑓𝑡= 𝑉 𝑅𝐸𝐹(𝑀𝐴𝑋) − 𝑉 𝑅𝐸𝐹(𝑀𝐼𝑁) 𝑉 𝑅𝐸𝐹 ∙𝛥𝑇 ∙ 10 6
Green
𝐷𝑟𝑖𝑓𝑡= − ∙(125°𝐶− −50°𝐶 ) ∙ 10 6 =5.7𝑝𝑝𝑚/°𝐶
Red
𝐷𝑟𝑖𝑓𝑡= − ∙(125𝐶− −50𝐶 ) ∙ 10 6 =20.5𝑝𝑝𝑚/°𝐶
Change in output voltage vs temp.
Box method: drift over the entire range.
Instantaneous drift can be higher or lower.
If example limit is 20ppm/⁰C, red fails and green passes.
Temperature drift is the change in output voltage over temperature. This drift is normally very nonlinear for references as it uses design techniques and trim to get a relatively flat temperature response, so the residual error is nonlinear. The box method is used to calculate temperature drift. This method looks for the largest change in output voltage over temperature. The drift in any particular region can be smaller or larger than the specification. In the example, you can see that the drift from 0C to 50C is relatively flat where as the drift at low and high temperature is much higher. The box spec gives the drift across the entire range. Also note that the drift is only tested at a few key points. In this example, it is tested at -50C, 25C, and 125C. It is possible that a minima or maxima could occur at temperatures other than the test points, but the test points were selected according to the expected curvature of the drift to avoid this. In this example the limit is set to 20ppm/C, so the green curve passes and the red curve fails the limits. Temperature drift is considered a very important specification because it is usually not practical to calibrate out this error. Typically, to achieve the best temperature drift you simply choose the product with the best drift specification.

10. 0.1Hz to 10Hz Noise: 1/f noise (flicker noise)
Measured using a special band pass filter and oscilloscope.
Not practical to filter 1/f.
Looks like temperature drift.
Good relative comparison of 1/f noise for different references.
0.1Hz to 10Hz noise is an industry standard way of specifying 1/f or flicker noise. It is measured using an active band pass filter that attenuates signals below 0.1Hz and above 10Hz. It is not practical to filter 1/f noise as it extends down to zero frequency and you cannot set a filter cutoff to zero hertz. Because this noise is very low in frequency it can be difficult to distinguish from ambient temperature drift. The only practical way to minimize 1/f noise in references is to simply choose a reference model with good specifications. Because this error source cannot be calibrated out it is often a key consideration when selecting devices.

11. Broadband noise minimized with filtering
1/f noise
Broadband noise minimized with filtering
Broadband noise
This plot shows the output noise spectrum of the reference.
[click] At low frequencies, the plot shows the 1/f noise component of the reference.
[click] At higher frequencies, the flat portion of the curve shows the broadband noise of the reference. Broadband noise can be minimized by filtering the reference. One approach is to simply add a larger filter capacitor at the output of the reference.
This graph shows how broadband noise is minimized using larger output capacitors.
[click] Another approach to minimizing broadband noise is to use an RC filter with a reference buffer. In this case it is possible to set the cutoff frequency very low to minimize the broadband noise. Note that this method is not practical without a reference buffer as the series resistance will limit the dynamic output impedance of the reference.

12. Output Impedance REF60xx
Low output impedance across frequency allows the reference to respond to rapid current transients.
One specification that is important to driving the SAR ADC reference input is the output impedance. Unfortunately, many references do not provide information on this specification. The graph shown here is for the REF60xx series reference. You can see in this example that the output impedance is very low out to 1MHz. This reference was specifically optimized to drive reference inputs so this output impedance curve is a good bench mark for comparison.

13. Important reference specification
Mitigation
Initial Error
Can be calibrated out.
Solder shift
Can be calibrated out. Can be minimized by carful PCB design and assembly.
Temperature Drift
Very difficult to calibrate.
Long Term Shift
Periodic calibrations can minimize this error. Also, accelerated aging (i.e. baking the system) can be used to eliminate the largest shifts.
Broadband Noise
Increasing the filtering can minimize broadband noise.
1/f Noise
Not practical to eliminate. Choose the lowest 0.1Hz to 10Hz device
Output impedance
Choose references with low output impedance across frequency for driving dynamic loads like SAR reference inputs.
This table summarizes the different reference error sources that we discussed. The key point to remember for each error source is how the error can be mitigated. The initial error and solder shift can be calibrated out. Temperature drift on the other hand, is difficult to calibrate out so this is usually minimized by selecting a device with specifications that meet your requirements. Long term shift, or aging, can be eliminated by periotic calibrations. Also, you can bake the product to accelerate aging and eliminate the largest shifts. Broadband noise can be minimized using filtering, but 1/f noise can only be minimized by selecting a device with good specifications. Finally, low wideband output impedance is important for applications with dynamic loads like SAR reference inputs. The only way to “mitigate” a non-optimal output impedance is to use a reference buffer. Otherwise, just select a device with low, wide bandwidth output impedance.

14. Thanks for your time! Please try the quiz.
That concludes this video – thank you for watching! Please try the quiz to check your understanding of this video’s content.

15. Leave on screen for 5 seconds.